#include "AP_Math.h" #include #include /* * is_equal(): Integer implementation, provided for convenience and * compatibility with old code. Expands to the same as comparing the values * directly */ template typename std::enable_if::type>::value ,bool>::type is_equal(const Arithmetic1 v_1, const Arithmetic2 v_2) { typedef typename std::common_type::type common_type; return static_cast(v_1) == static_cast(v_2); } /* * is_equal(): double/float implementation - takes into account * std::numeric_limits::epsilon() to return if 2 values are equal. */ template typename std::enable_if::type>::value, bool>::type is_equal(const Arithmetic1 v_1, const Arithmetic2 v_2) { #ifdef ALLOW_DOUBLE_MATH_FUNCTIONS typedef typename std::common_type::type common_type; typedef typename std::remove_cv::type common_type_nonconst; if (std::is_same::value) { return fabs(v_1 - v_2) < std::numeric_limits::epsilon(); } #endif #pragma clang diagnostic push #pragma clang diagnostic ignored "-Wabsolute-value" // clang doesn't realise we catch the double case above and warns // about loss of precision here. return fabsf(v_1 - v_2) < std::numeric_limits::epsilon(); #pragma clang diagnostic pop } template bool is_equal(const int v_1, const int v_2); template bool is_equal(const short v_1, const short v_2); template bool is_equal(const long v_1, const long v_2); template bool is_equal(const float v_1, const float v_2); template bool is_equal(const double v_1, const double v_2); template float safe_asin(const T v) { const float f = static_cast(v); if (isnan(f)) { return 0.0f; } if (f >= 1.0f) { return static_cast(M_PI_2); } if (f <= -1.0f) { return static_cast(-M_PI_2); } return asinf(f); } template float safe_asin(const int v); template float safe_asin(const short v); template float safe_asin(const float v); template float safe_asin(const double v); template float safe_sqrt(const T v) { float ret = sqrtf(static_cast(v)); if (isnan(ret)) { return 0; } return ret; } template float safe_sqrt(const int v); template float safe_sqrt(const short v); template float safe_sqrt(const float v); template float safe_sqrt(const double v); /* replacement for std::swap() needed for STM32 */ static void swap_float(float &f1, float &f2) { float tmp = f1; f1 = f2; f2 = tmp; } /* * linear interpolation based on a variable in a range */ float linear_interpolate(float low_output, float high_output, float var_value, float var_low, float var_high) { if (var_low > var_high) { // support either polarity swap_float(var_low, var_high); swap_float(low_output, high_output); } if (var_value <= var_low) { return low_output; } if (var_value >= var_high) { return high_output; } float p = (var_value - var_low) / (var_high - var_low); return low_output + p * (high_output - low_output); } /* cubic "expo" curve generator * alpha range: [0,1] min to max expo * input range: [-1,1] */ float expo_curve(float alpha, float x) { return (1.0f - alpha) * x + alpha * x * x * x; } /* throttle curve generator * thr_mid: output at mid stick * alpha: expo coefficient * thr_in: [0-1] */ float throttle_curve(float thr_mid, float alpha, float thr_in) { float alpha2 = alpha + 1.25 * (1.0f - alpha) * (0.5f - thr_mid) / 0.5f; alpha2 = constrain_float(alpha2, 0.0f, 1.0f); float thr_out = 0.0f; if (thr_in < 0.5f) { float t = linear_interpolate(-1.0f, 0.0f, thr_in, 0.0f, 0.5f); thr_out = linear_interpolate(0.0f, thr_mid, expo_curve(alpha, t), -1.0f, 0.0f); } else { float t = linear_interpolate(0.0f, 1.0f, thr_in, 0.5f, 1.0f); thr_out = linear_interpolate(thr_mid, 1.0f, expo_curve(alpha2, t), 0.0f, 1.0f); } return thr_out; } template T wrap_180(const T angle) { auto res = wrap_360(angle); if (res > T(180)) { res -= T(360); } return res; } template T wrap_180_cd(const T angle) { auto res = wrap_360_cd(angle); if (res > T(18000)) { res -= T(36000); } return res; } template int wrap_180(const int angle); template short wrap_180(const short angle); template float wrap_180(const float angle); #ifdef ALLOW_DOUBLE_MATH_FUNCTIONS template double wrap_180(const double angle); #endif template int wrap_180_cd(const int angle); template long wrap_180_cd(const long angle); template short wrap_180_cd(const short angle); template float wrap_180_cd(const float angle); #ifdef ALLOW_DOUBLE_MATH_FUNCTIONS template double wrap_180_cd(const double angle); #endif float wrap_360(const float angle) { float res = fmodf(angle, 360.0f); if (res < 0) { res += 360.0f; } return res; } #ifdef ALLOW_DOUBLE_MATH_FUNCTIONS double wrap_360(const double angle) { double res = fmod(angle, 360.0); if (res < 0) { res += 360.0; } return res; } #endif int wrap_360(const int angle) { int res = angle % 360; if (res < 0) { res += 360; } return res; } float wrap_360_cd(const float angle) { float res = fmodf(angle, 36000.0f); if (res < 0) { res += 36000.0f; } return res; } #ifdef ALLOW_DOUBLE_MATH_FUNCTIONS double wrap_360_cd(const double angle) { double res = fmod(angle, 36000.0); if (res < 0) { res += 36000.0; } return res; } #endif int wrap_360_cd(const int angle) { int res = angle % 36000; if (res < 0) { res += 36000; } return res; } long wrap_360_cd(const long angle) { long res = angle % 36000; if (res < 0) { res += 36000; } return res; } ftype wrap_PI(const ftype radian) { ftype res = wrap_2PI(radian); if (res > M_PI) { res -= M_2PI; } return res; } ftype wrap_2PI(const ftype radian) { ftype res = fmodF(radian, M_2PI); if (res < 0) { res += M_2PI; } return res; } template T constrain_value_line(const T amt, const T low, const T high, uint32_t line) { // the check for NaN as a float prevents propagation of floating point // errors through any function that uses constrain_value(). The normal // float semantics already handle -Inf and +Inf if (isnan(amt)) { AP::internalerror().error(AP_InternalError::error_t::constraining_nan, line); return (low + high) / 2; } if (amt < low) { return low; } if (amt > high) { return high; } return amt; } template float constrain_value_line(const float amt, const float low, const float high, uint32_t line); template double constrain_value_line(const double amt, const double low, const double high, uint32_t line); template T constrain_value(const T amt, const T low, const T high) { // the check for NaN as a float prevents propagation of floating point // errors through any function that uses constrain_value(). The normal // float semantics already handle -Inf and +Inf if (std::is_floating_point::value) { if (isnan(amt)) { INTERNAL_ERROR(AP_InternalError::error_t::constraining_nan); return (low + high) / 2; } } if (amt < low) { return low; } if (amt > high) { return high; } return amt; } template int constrain_value(const int amt, const int low, const int high); template long constrain_value(const long amt, const long low, const long high); template long long constrain_value(const long long amt, const long long low, const long long high); template short constrain_value(const short amt, const short low, const short high); template float constrain_value(const float amt, const float low, const float high); template double constrain_value(const double amt, const double low, const double high); /* simple 16 bit random number generator */ uint16_t get_random16(void) { static uint32_t m_z = 1234; static uint32_t m_w = 76542; m_z = 36969 * (m_z & 0xFFFFu) + (m_z >> 16); m_w = 18000 * (m_w & 0xFFFFu) + (m_w >> 16); return ((m_z << 16) + m_w) & 0xFFFF; } #if CONFIG_HAL_BOARD == HAL_BOARD_SITL // generate a random float between -1 and 1, for use in SITL float rand_float(void) { return ((((unsigned)random()) % 2000000) - 1.0e6) / 1.0e6; } Vector3f rand_vec3f(void) { Vector3f v = Vector3f(rand_float(), rand_float(), rand_float()); if (!is_zero(v.length())) { v.normalize(); } return v; } #endif /* return true if two rotations are equivalent This copes with the fact that we have some duplicates, like ROLL_180_YAW_90 and PITCH_180_YAW_270 */ bool rotation_equal(enum Rotation r1, enum Rotation r2) { if (r1 == r2) { return true; } Vector3f v(1,2,3); Vector3f v1 = v; Vector3f v2 = v; v1.rotate(r1); v2.rotate(r2); return (v1 - v2).length() < 0.001; } /* * return a velocity correction (in m/s in NED) for a sensor's position given it's position offsets * this correction should be added to the sensor NED measurement * sensor_offset_bf is in meters in body frame (Foward, Right, Down) * rot_ef_to_bf is a rotation matrix to rotate from earth-frame (NED) to body frame * angular_rate is rad/sec */ Vector3F get_vel_correction_for_sensor_offset(const Vector3F &sensor_offset_bf, const Matrix3F &rot_ef_to_bf, const Vector3F &angular_rate) { if (sensor_offset_bf.is_zero()) { return Vector3F(); } // correct velocity const Vector3F vel_offset_body = angular_rate % sensor_offset_bf; return rot_ef_to_bf.mul_transpose(vel_offset_body) * -1.0; } /* calculate a low pass filter alpha value */ float calc_lowpass_alpha_dt(float dt, float cutoff_freq) { if (dt <= 0.0f || cutoff_freq <= 0.0f) { return 1.0; } float rc = 1.0f/(M_2PI*cutoff_freq); return constrain_float(dt/(dt+rc), 0.0f, 1.0f); } #if CONFIG_HAL_BOARD == HAL_BOARD_SITL // fill an array of float with NaN, used to invalidate memory in SITL void fill_nanf(float *f, uint16_t count) { const float n = std::numeric_limits::signaling_NaN(); while (count--) { *f++ = n; } } void fill_nanf(double *f, uint16_t count) { while (count--) { *f++ = std::numeric_limits::signaling_NaN(); } } #endif // Convert 16-bit fixed-point to float float fixed2float(const uint16_t input, const uint8_t fractional_bits) { return ((float)input / (float)(1U << fractional_bits)); } // Convert float to 16-bit fixed-point uint16_t float2fixed(const float input, const uint8_t fractional_bits) { return (uint16_t)(roundf(input * (1U << fractional_bits))); } /* calculate turn rate in deg/sec given a bank angle and airspeed for a fixed wing aircraft */ float fixedwing_turn_rate(float bank_angle_deg, float airspeed) { bank_angle_deg = constrain_float(bank_angle_deg, -80, 80); return degrees(GRAVITY_MSS*tanf(radians(bank_angle_deg))/MAX(airspeed,1)); }